Tool Wear Volume and Residual Life Prediction Based on Extreme Learning Machine

Abstract

Since the tool wear state has a great influence on processing quality, an online monitoring method for tool wear state based on single-hidden layer feed-forward neural network (SLFN)—extreme learning machine (ELM) is proposed. According to the real-time working condition data of the milling tool, a variety of feature extraction methods including statistical analysis, fast Fourier transform and wavelet transform are used to extract 34 features. The 34 features are mixed features of time domain, frequency domain and time-frequency domain and they are sensitive to tool wear. The extracted features and wear volume are input into the extreme learning machine network framework for training to get the ELM model. Then put the test set features into the ELM model for prediction and evaluation, and the corresponding predicted tool wear volume and remaining number of passes are obtained. By comparing the prediction results of SVR and BPNN optimized by genetic algorithm, it is found that ELM has the advantages of fast learning speed, high prediction accuracy and strong generalization ability, and can realize online monitoring of tool wear state.